Dirichlet’s Lemma in Number Fields Franz Lemmermeyer Abstract . Dirichlet’s Lemma states that every primitive quadratic Dirichlet character χ \chi can be written in the form χ ⁡ ( n ) = ( Δ n ) \chi(n)=(\frac{\Delta}{n}) for a suitable quadratic discriminant Δ \Delta . In this article we define a group, the separant class group SCl ​ ( F ) \mbox{\rm SCl}(F) , that measures the extent to which Dirichlet’s Lemma fails in general number fields F F . As an application we will show that over fields with trivial separant groups, genus theory of quadratic extensions can be made as explicit as over the rationals. In the first part of this article we will introduce the separant class group of a number field F as a finite 2 -group that is trivial if and only if every quadratic Dirichlet character can be realized as a Kronecker symbol. The second part is devoted to applications. 1. Genus Theory of Quadratic Number Fields We begin by reviewing a few classical results concerning quadratic number fields, the factorization of discriminants into prime discriminants, and connections with genus theory. Let \Delta denote the discriminant of a quadratic number field K . Quadratic discriminants form a group {\operatorname{Disc}\,} with respect to the following multiplication: if \Delta_{1} and \Delta_{2} are two quadratic discriminants, then \Delta_{1}*\Delta_{2}=\Delta_{3} , where \Delta_{3} is the discriminant of the extension {\mathbb{Q}}(\sqrt{\Delta_{1}\Delta_{2}}\,) . The neutral element is the discriminant 1 of the trivial extension {\mathbb{Q}}/{\mathbb{Q}} . Observe that \Delta*\Delta=1 (the group {\operatorname{Disc}\,} is an elementary abelian 2 -group) and that, e.g., (-8)*(-4)=8 . If \Delta_{1} and \Delta_{2} are coprime, then clearly \Delta_{1}*\Delta_{2}=\Delta_{1}\Delta_{2} . Prime discriminants are discriminants having only one prime factor, namely • the discriminants -4 , \pm 8 of {\mathbb{Q}}(\sqrt{-1}\,) and {\mathbb{Q}}(\sqrt{\pm 2}\,) ; • the discriminant p of {\mathbb{Q}}(\sqrt{p}\,) for primes p\equiv 1\bmod 4 ; • the discriminant -q of {\mathbb{Q}}(\sqrt{-q}\,) for primes q\equiv 3\bmod 4 . It is well known and easy to show that every discriminant \Delta can be written uniquely as a product of prime discriminants: \Delta=\Delta_{1}\cdots\Delta_{r}\Delta_{r+1}\cdots\Delta_{t}, where we assume that \Delta_{j}<0 for j\leq r and \Delta_{j}>0 for j>r . In this case, K_{\operatorname{gen}}^{+}={\mathbb{Q}}(\sqrt{\Delta_{1}},\ldots,\sqrt{\Delta_{t}}\,) is the genus class field of K in the strict sense, that is, the maximal abelian extension (necessarily an elementary abelian 2 -extension) of {\mathbb{Q}} that is unramified at all finite places over K . The ordinary genus field K_{\operatorname{gen}} of K , i.e., the maximal unramified extension of K that is abelian over {\mathbb{Q}} , is then given by K_{\operatorname{gen}}={\mathbb{Q}}(\sqrt{\Delta_{1}\Delta_{2}},\ldots,\sqrt{\Delta_{1}\Delta_{r}},\sqrt{\Delta_{r+1}},\ldots,\sqrt{\Delta_{t}}\,) if \Delta>0 (for negative discriminants, K_{\operatorname{gen}}=K_{\operatorname{gen}}^{+} ). Example 1. The quadratic number field K={\mathbb{Q}}(\sqrt{210}\,) has discriminant \Delta=840 , and the factorization of \Delta into prime discriminants is \Delta=8\cdot(-3)\cdot 5\cdot(-7), hence the genus field of K in the usual and in the strict sense are given by K_{\operatorname{gen}}^{+}={\mathbb{Q}}(\sqrt{2},\sqrt{-3},\sqrt{5},\sqrt{-7}\,)\quad\text{and}\quad K_{\operatorname{gen}}={\mathbb{Q}}(\sqrt{2},\sqrt{5},\sqrt{21}\,). The genus field K_{\operatorname{gen}} in the usual sense is the maximal real subfield of K_{\operatorname{gen}}^{+} , which in our case is K_{\operatorname{gen}}={\mathbb{Q}}(\sqrt{2},\sqrt{5},\sqrt{21}\,). A quadratic number field has odd class number (in the usual sense) if and only if K_{\operatorname{gen}}=K . This condition is obviously necessary; for sufficiency observe that if K has even class number, then its Hilbert 2 -class field can be constructed in central steps, i.e., there exists a central unramified quadratic extension of K , which necessarily is elementary abelian over {\mathbb{Q}} and thus contained in the genus class field of K (for details, see [ 5 ] ). The explicit description of the genus field then immediately implies the following classification of quadratic number fields with odd class number: Theorem 1.1 . The quadratic extensions {\mathbb{Q}}(\sqrt{m}\,) with odd class number are • {\mathbb{Q}}(\sqrt{-1}\,) , {\mathbb{Q}}(\sqrt{-2}\,) , and {\mathbb{Q}}(\sqrt{-q}\,) for primes q\equiv 3\bmod 4 in the complex quadratic case; • {\mathbb{Q}}(\sqrt{p}\,) for primes p and {\mathbb{Q}}(\sqrt{pq}\,) , for primes p and q , where p=2 or p\equiv 3\bmod 4 and q\equiv 3\bmod 4 . Genus characters Given an ideal class c\in\operatorname{Cl}^{+}(K) , choose an ideal {\mathfrak{a}} coprime to \Delta . Then \chi_{j}(c)=(\frac{\Delta_{j}}{N{\mathfrak{a}}}) defines a quadratic character on the ideal class group \operatorname{Cl}^{+}(K) called a genus character. The main result concerning genus characters is the Theorem 1.2 (Principal Genus Theorem) . An ideal class c\in\operatorname{Cl}^{+}(K) is a square if and only if \chi_{j}(c)=1 for j=1,\ldots,t . Work by Goldstein [ 6 ] and Sunley [ 12 , 13 ] (see also Davis [ 3 ] ) shows that, for a base field F , unique factorization into prime discriminants holds if and only if F is totally real and has class number 1 in the strict sense. Replacing discriminants by separants we will be able to generalize these results to fields with trivial separant class group, i.e., to totally real fields with odd class number in the strict sense. The notion of separants goes back to the author’s dissertation [ 8 ] , where it was used to generalize Rédei’s construction of unramified cyclic quartic extensions. 2. Dirichlet’s Lemma in {\mathbb{Q}} In this article we will study Kronecker symbols (\frac{\Delta}{p}) , which describe the splitting of primes in the quadratic number field with discriminant \Delta , and quadratic Dirichlet characters defined modulo m : these are homomorphisms from ({\mathbb{Z}}/m{\mathbb{Z}})^{\times} to the (multiplicatively written) group {\mathbb{Z}}/2{\mathbb{Z}} . We identify Dirichlet characters modulo m and modulo n if they agree on all integers coprime to mn . A quadratic Dirichlet character defined modulo m is called primitive if it cannot be defined for any proper divisor of n . The three nontrivial quadratic Dirichlet characters defined modulo 8 are \begin{array}[]{c|rrrr}\raisebox{0.0pt}[11.19443pt][4.30554pt]{\rule{0.0pt}{8.61108pt}}&1&3&5&7\\ \hline\cr\raisebox{0.0pt}[11.19443pt][4.30554pt]{\rule{0.0pt}{8.61108pt}}\chi_{8}&+1&-1&-1&+1\\ \raisebox{0.0pt}[11.19443pt][4.30554pt]{\rule{0.0pt}{8.61108pt}}\chi_{-8}&+1&+1&-1&-1\\ \raisebox{0.0pt}[11.19443pt][4.30554pt]{\rule{0.0pt}{8.61108pt}}\chi_{-4}&+1&-1&+1&-1\end{array} Their values at positive integers agree with (\frac{2}{p}) , (\frac{-2}{p}) and (\frac{-1}{p}) , respectively. The characters \chi_{8} and \chi_{-8} are primitive, the character \chi_{-4} is not as it is already defined modulo 4 . The smallest defining modulus of a character is called its conductor; for Kronecker characters \kappa_{a}=(\frac{a}{\cdot}) , the conductor agrees with the discriminant of the quadratic number field {\mathbb{Q}}(\sqrt{a}\,) . The fact that Kronecker symbols (\frac{\Delta}{p}) for primes p>0 only depend on the residue class of p modulo \Delta is essentially equivalent to the quadratic reciprocity law. This observation assigns a primitive quadratic Dirichlet character to each Kronecker symbol, and Dirichlet’s Lemma states that every primitive quadratic Dirichlet character arises in this way (see below for a precise statement). The same construction works for any algebraic number field, but it turns out that, in general, not every primitive quadratic Dirichlet character comes from a Kronecker symbol. For measuring this defect we introduce the separant class group \mbox{\rm SCl}(F) of a number field F , and we will determine its structure. We start by recalling Dirichlet’s Lemma 1 1 1 The name “Dirichlet’s Lemma” was coined by Cohn [ 2 ] . , which goes back to Dirichlet’s proof that there are infinitely many primes in arithmetic progressions. In this proof, Dirichlet invented the characters that now bear his name (strictly speaking Dirichlet provided the tools, and Dedekind came up with the abstract concept of a character when he edited Dirichlet’s lectures). Dirichlet characters are group homomorphisms \chi:({\mathbb{Z}}/m{\mathbb{Z}})^{\times}\longrightarrow{\mathbb{C}}^{\times} . Dirichlet’s main problem was showing that L-series L(s,\chi)=\sum_{n=1}^{\infty}\chi(n)n^{-s} associated to Dirichlet characters \chi have the property (1) \lim_{s\to 1}L(s,\chi)\neq 0 for all characters \chi different from the principal character. This was not difficult except for quadratic characters \chi , i.e., characters that attain only the values +1 and -1 . Eventually, Dirichlet realized that the limit in ( 1 ) could be evaluated directly by showing that every primitive quadratic Dirichlet character \chi has the form \chi(n)=(\frac{\Delta}{n}) for some quadratic discriminant \Delta , which allowed him to show that the limit in question is a product of nonzero constants and the class number and the regulator of the quadratic number field {\mathbb{Q}}(\sqrt{D}\,) . Again this is only true cum grano salis, as Dirichlet worked with quadratic forms instead of fields. Let us call a character \kappa of the form \kappa(n)=(\frac{\Delta}{n}) a Kronecker character. A Kronecker character \kappa is called modular if there exists a modulus N>0 such that \kappa(n)=\kappa(n+N) for all natural numbers n . It is easy to see that if \kappa is modular with respect to N_{1} and N_{2} , then it is modular with respect to \gcd(N_{1},N_{2}) . The minimal modulus is called the conductor of \kappa . Similarly, a quadratic number field K is called modular if there exists an integer N>0 such that K is contained in the field {\mathbb{Q}}(\zeta_{N}) of N -th roots of unity. Again it is easy to see that if K is modular with respect to N_{1} and N_{2} , then it is modular with respect to \gcd(N_{1},N_{2}) . The minimal modulus for K is called the conductor of K . The similarity between these two notions of modularity is not superficial: the splitting of primes p in quadratic number fields K with discriminant \Delta depends on the Kronecker character \kappa(p)=(\frac{\Delta}{p}) , that in cyclotomic fields {\mathbb{Q}}(\zeta_{N}) on the values \chi(p) for certain Dirichlet characters \chi defined modulo N . The modularity of K can be shown to imply the modularity of \kappa , and in fact the conductor of the Kronecker character (\frac{\Delta}{\cdot}) is equal to the conductor N=|\Delta| of the corresponding quadratic number field K . Infinite Kronecker characters Consider the Kronecker symbol \kappa_{4}(n)=(\frac{-4}{n}) . Then for all positive integers n we have \kappa_{4}(n)=\begin{cases}+1&\text{ if }n\equiv 1\bmod 4,\\ -1&\text{ if }n\equiv 3\bmod 4.\end{cases} There is, however, a problem with negative integers: although -3\equiv 1\bmod 4 , we have \kappa_{4}(-3)=\kappa_{4}(+3)=+1 . To get around this problem we introduce an infinite Kronecker symbol \kappa_{\infty}(n)=\Big(\frac{\infty}{n}\Big)=\begin{cases}+1&\text{ if }n>0,\\ -1&\text{ if }n<0.\end{cases} Then \kappa_{4}(n)\kappa_{\infty}(n)=\chi(n) for all nonzero integers n , where \chi is the nontrivial Dirichlet character modulo 4 . Dirichlet showed that every Dirichlet character is a Kronecker character; in fact, we even have Dirichlet’s Lemma. There is a bijection between primitive quadratic Dirichlet characters modulo N and Kronecker characters with conductor N . The fact that, conversely, Kronecker characters (\frac{\Delta}{\cdot}) are Dirichlet characters defined modulo |\Delta| is equivalent to Euler’s version of the quadratic reciprocity law (see [ 1 ] , as well as, for a proof of Dirichlet’s Lemma, [ 10 ] .). 3. Dirichlet’s Lemma in Number Fields We now define the groups of Kronecker characters {\operatorname{Kr}}(F) and the group of quadratic Dirichlet characters {\operatorname{Dir}}(F) in arbitrary number fields F , observe that each Kronecker character is a Dirichlet character, and define the separant class group as their quotient group. Kronecker characters Let F be a number field. Each element \alpha\in F^{\times} defines a Kronecker character \kappa_{\alpha}=(\frac{\alpha}{}) . We identify two Kronecker characters \kappa_{\alpha} and \kappa_{\beta} if they have the same values at all elements coprime to \alpha and \beta ; for example, we identify the Kronecker symbols (\frac{-1}{\cdot}) and (\frac{-4}{\cdot}) over {\mathbb{Q}} . We will evaluate Kronecker characters (\frac{\alpha}{\beta}) only at integers \beta coprime to the relative discriminant \operatorname{disc}\,(K/F) , where K=F(\sqrt{\alpha}\,) . For example, the Kronecker symbol (\frac{3}{m}) is defined only for integers coprime to 6 . Kronecker characters form a group {\operatorname{Kr}}(F) with respect to multiplication. If \alpha is a square, \kappa_{\alpha} is the trivial character. This observation implies that there is a homomorphism \phi:F^{\times}/F^{\times\,2}\longrightarrow{\operatorname{Kr}}(F) defined by \phi(\alpha F^{\times\,2})=\kappa_{\alpha} . Different elements in F^{\times}/F^{\times\,2} do not necessarily give rise to different Kronecker symbols; in other words: \phi is not necessarily injective. The next proposition determnines the kernel of \phi : Proposition 3.1 . We have an exact sequence \begin{CD}1@>{}>{}>E_{4}/E^{2}@>{\iota}>{}>F^{\times}/F^{\times\,2}@>{\phi}>{}>{\operatorname{Kr}}(F)@>{}>{}>1,\end{CD} where E is the unit group of {\mathcal{O}}_{F} and E_{4} is the subgroup of units that are congruent to a square modulo 4 . Observe that if \varepsilon\in E_{4}\setminus E^{2} , then F(\sqrt{\varepsilon}\,)/F is a quadratic extension unramified at all the finite primes; in particular, the class number of F in the strict sense is even when such a unit exists. An explicit example of a unit in E_{4}\setminus E^{2} is given by the fundamental unit \varepsilon=35+6\sqrt{34} of {\mathbb{Q}}(\sqrt{34}\,) ; since 2\varepsilon=(6+\sqrt{34}\,)^{2} we have F(\sqrt{\varepsilon}\,)=F(\sqrt{2}\,) , and this is an unramified quadratic extension of F . Proof. We only have to show that \ker\phi=\operatorname{im}\iota . Assume that \kappa\in\ker\phi ; then \kappa=(\frac{\alpha}{}) is the trivial Kronecker character. Since the conductor of the Kronecker symbol is the conductor of the quadratic extension F(\sqrt{\alpha})/F , this extension is unramified at all finite primes; but this implies that either \alpha is a square and the extension is trivial, or that \alpha is a unit congruent to a square modulo 4 . ∎ Next we show that Kronecker characters are Dirichlet characters: Proposition 3.2 . Each Kronecker character \kappa_{\alpha}=(\frac{\alpha}{\cdot}) is a primitive quadratic Dirichlet character with conductor \operatorname{disc}\,(K/F) , where K=F(\sqrt{\alpha}\,) . Proof. By Artin’s reciprocity law, the splitting behavior of a prime ideal in a quadratic extension K=F(\sqrt{\alpha}\,) only depends on the coset of {\mathfrak{p}} in the corresponding ideal group, and the conductor of this group is the relative discriminant of K/F . In particular, for any \beta\gg 0 in F^{\times} coprime to 2\alpha we have (\frac{\alpha}{\beta})=\chi(\beta) for some Dirichlet character with conductor \operatorname{disc}\,(K/F) . Since the conductor is the minimal defining modulus, \chi is primitive. ∎ This result will allow us to show that certain Dirichlet characters are not Kronecker characters: a necessary condition for the existence if a Kronecker character with conductor {\mathfrak{f}} is the existence of a quadratic extension K/F with relative discriminant \operatorname{disc}\,(K/F)={\mathfrak{f}} . 4. Determination of the Separant Class Group The separant class group \mbox{\rm SCl}(F)={\operatorname{Dir}}(F)/{\operatorname{Kr}}(F) of a number field F is by definition an elementary abelian 2 -group, and so it is sufficient to determine its order, or its dimension as an {\mathbb{F}}_{2} -vector space. This is accomplished by our main theorem: Theorem 4.1 . There is an isomorphism \mbox{\rm SCl}(F)\simeq\operatorname{Cl}_{F}\{4\}/\operatorname{Cl}_{F}\{4\}^{2}, where \operatorname{Cl}_{F}\{4\} is the ray class group of F defined modulo (4) . In particular we have \#\mbox{\rm SCl}(F)=2^{\rho^{+}+s} , where 2^{\rho^{+}}=\operatorname{Cl}_{2}^{+}(K)/\operatorname{Cl}_{2}^{+}(K)^{2} denotes the 2 -part of the class group of F in the strict sense, and s the number of pairs of complex embeddings of F . Thus F has trivial separant class group if and only if F is totally real with odd class number in the strict sense. The fact that \#\mbox{\rm SCl}(F)=2^{\rho^{+}+s} was proved in [ 9 ] . Characters with conductor dividing (4) Our proof uses a few facts concerning Kronecker and Dirichlet characters whose conductors divide (4) . Let {\operatorname{Kr}}_{4}(F) denote the group of Kronecker characters with conductor dividing (4) , {\operatorname{Dir}}_{4}(F) the group of Dirichlet characters defined modulo (4) , and M_{4}=({\mathcal{O}}_{F}/4{\mathcal{O}}_{F})^{\times} the group of coprime residue classes modulo 4 . For the proof of Thm. 4.1 we only need the fact that the homomorphism \iota_{2}:M_{4}/M_{4}^{2}\longrightarrow{\operatorname{Dir}}_{4}(F) is surjective. Proposition 4.2 . There is an isomorphism \iota_{2}:M_{4}/M_{4}^{2}\longrightarrow{\operatorname{Dir}}_{4}(F) . Both groups have order 2^{n} , where n=(F:{\mathbb{Q}}) . Proof. Quadratic Dirichlet characters defined modulo 4 are homomorphisms from M_{4} to the multiplicatively written group {\mathbb{Z}}/2{\mathbb{Z}} . Since such homomorphisms are trivial on squares, they induce homomorphisms M_{4}/M_{4}^{2}\longrightarrow{\mathbb{Z}}/2{\mathbb{Z}} . Thus {\operatorname{Dir}}_{4}(F)=\operatorname{Hom}(M_{4}/M_{4}^{2},{\mathbb{Z}}/2{\mathbb{Z}}) , which means that {\operatorname{Dir}}_{4}(F) is the dual of M_{4}/M_{4}^{2} (for example as {\mathbb{Z}}/2{\mathbb{Z}} -vector spaces); in particular, {\operatorname{Dir}}_{4}(F)\simeq M_{4}/M_{4}^{2} as abelian groups. The isomorphism between M_{4}/M_{4}^{2} and {\operatorname{Dir}}_{4}(F) given above is not canonical, but depends on the choice of a basis. In this case, however, there is also a canonical isomorphism that we will now construct (and which we need for making the diagram commutative). Let \alpha represent a coprime residue class modulo 4 , and let \kappa_{\alpha} denote the corresponding Kronecker character. Write \kappa_{\alpha}=\chi_{0}\chi_{1} as the product of a primitive quadratic Dirichlet character \chi_{0} defined modulo 4 and a Dirichlet character \chi_{1} defined modulo an odd ideal. Then set \iota_{2}(\alpha E^{2})=\chi_{0} . The map \iota_{2} is well defined: If \alpha\equiv\beta\bmod 4 , \kappa_{\alpha}=\chi_{0}\chi_{1} and \kappa_{\beta}=\chi_{0}^{\prime}\chi_{1}^{\prime} , then \kappa_{\alpha}\kappa_{\beta}=\kappa_{\alpha\beta} is defined modulo an odd ideal since \alpha/\beta\equiv 1\bmod 4 . Thus \chi_{0}=\chi_{0}^{\prime} . We observe that if \alpha is a unit, then \iota_{2}(\alpha E^{2})=\kappa_{\varepsilon} ; thus the upper left square of the diagram is commutative. We claim that \iota_{2} is injective. Assume that \iota_{2}(\alpha M_{4}^{2})=\iota_{2}(\beta M_{4}^{2}) . Then \kappa_{\alpha\beta} has conductor coprime to 2 , hence \alpha\beta\equiv\xi^{2}\bmod 4 , which implies that \alpha and \beta generate the same class in M_{4}/M_{4}^{2} . Since M_{4}/M_{4}^{2}\simeq({\mathbb{Z}}/2{\mathbb{Z}})^{n} for n=(F:{\mathbb{Q}}) (see [ 9 , p. 281] ; this follows by sending \alpha\bmod 2 to 1+2\alpha\bmod 4 ), both groups have order 2^{n} . ∎ Kronecker characters with conductor dividing (4) Next we determine the number of Kronecker characters with conductor dividing (4) . Recall that we have introduced the following groups in [ 9 ] : • \operatorname{Sel}(F)=\{\alpha\in F^{\times}:(\alpha)={\mathfrak{a}}^{2}\}/F^{\times\,2} is the group of all nonzero elements of F that generate the square of an ideal; • \operatorname{Sel}_{4}(F)=\{\alpha\in F^{\times}:(\alpha,2)=(1),(\alpha)={\mathfrak{a}}^{2},\alpha\equiv\xi^{2}\bmod 4\}/F^{\times\,2} is the subgroup of \operatorname{Sel}(F) of elements with odd norm that generate the square of an ideal and are congruent to a square modulo 4 . Proposition 4.3 . We have an exact sequence \begin{CD}1@>{}>{}>\operatorname{Sel}_{4}(F)@>{\iota}>{}>\operatorname{Sel}(F)@>{\nu}>{}>{\operatorname{Kr}}_{4}(F)@>{}>{}>1\end{CD} Proof. The homomorphism \iota is an injection since \operatorname{Sel}_{4}(F)\subseteq\operatorname{Sel}(F) . We set \nu(\alpha F^{\times\,2})=\kappa_{\alpha} . Then \ker\nu consists of the cosets \alpha F^{\times\,2} such that \kappa_{\alpha}=1 ; this holds if and only if \alpha\equiv\xi^{2}\bmod 4 , i.e., if \alpha F^{\times\,2}\in\operatorname{Sel}_{4}(F) . Finally assume that \kappa_{\alpha}\in{\operatorname{Kr}}_{4}(F) . Then (\alpha)={\mathfrak{a}}^{2} , hence \kappa_{\alpha}\in\operatorname{im}\nu . ∎ We know from [ 9 ] that \#\operatorname{Sel}_{4}(F)=2^{\rho^{+}} , where \rho^{+} denotes the 2 -rank of \operatorname{Cl}_{2}^{+}(F) , and that \#\operatorname{Sel}(F)=\rho+r+s , where \rho denotes the 2 -rank of \operatorname{Cl}_{2}(F) , and r and 2s denote the number of real and complex embeddings of F . Corollary 4.4 . We have \#{\operatorname{Kr}}_{4}(F)=2^{r+s-(\rho^{+}-\rho)} . In addition we see Corollary 4.5 . Every Dirichlet character defined modulo 4 is represented by a Kronecker character if and only if F is totally real and every totally positive unit in E_{F} is a square. The groups {\operatorname{Dir}}_{4}(F) and {\operatorname{Kr}}_{4}(F) have equal order if and only if n=r+s-(\rho^{+}-\rho) . Since n=r+2s , this is equivalent to s=-(\rho^{+}-\rho) . This in turn holds if and only if s=0 and \rho^{+}=\rho , i.e., if F is totally real and every totally positive unit is a square. Corollary 4.6 . If F has odd class number, then every Kronecker character with conductor dividing 4 has the form \kappa_{\varepsilon} for a unit \varepsilon\in E_{F} . Proof. If F has odd class number, then \operatorname{Sel}(F)\simeq E/E^{2} and \operatorname{Sel}_{4}(F)\simeq E_{4}/E^{2} by [ 9 , Prop. 3.1] . Thus we get the exact sequence \begin{CD}1@>{}>{}>E_{4}/E^{2}@>{\iota}>{}>E/E^{2}@>{\nu}>{}>{\operatorname{Kr}}_{4}(F)@>{}>{}>1,\end{CD} which implies the claim. ∎ In the case where the class number of F is odd, we may think of the elements in {\operatorname{Dir}}(F) as “ideal separants”: Proposition 4.7 . If F has odd class number in the strict sense, then Kronecker characters can be identified with separants of quadratic extensions K/F . It is not enough to assume that F has odd class number in the usual sense. In fact, consider the field F={\mathbb{Q}}(\sqrt{6}) with fundamental unit \varepsilon=5+2\sqrt{6} . The separant -\varepsilon E^{2} belongs to the quadratic extension K=F(\sqrt{-\varepsilon}\,)={\mathbb{Q}}(\sqrt{-2},\sqrt{-3}\,) since -\varepsilon=(\sqrt{-2}+\sqrt{-3}\,)^{2} , and K/F is unramified at all finite primes. This implies that principal ideals are in the kernel of the Artin map, which in turn means that the Kronecker symbol (\frac{-\varepsilon}{\cdot}) is trivial on F^{\times} . Proof. The natural homomorphism \kappa:F^{\times}/F^{\times\,2}\longrightarrow{\operatorname{Kr}}(F) defined by \kappa(\alpha F^{\times\,2})=(\frac{\alpha}{\cdot}) is clearly surjective. The kernel consists of all \alpha F^{\times\,2} such that \kappa(\alpha F^{\times\,2}) has conductor (1) . This is only possible if \alpha is (up to squares) a unit congruent to a square modulo 4 . Since F has odd class number in the strict sense, the only such unit is 1 . The separant \delta E^{2} corresponds to the Kronecker symbol (\frac{\delta}{\cdot}) . The homomorphism \lambda:\mbox{\rm Sep}(F)\longrightarrow{\operatorname{Kr}}(F) with \lambda(\delta E^{2})=\kappa_{\delta} is injective by what we have already shown. We claim that it is also surjective. In fact, let \kappa_{\alpha} denote any Kronecker character, and let \delta={\operatorname{sep}}(F(\sqrt{\alpha}\,)/F) . Then \delta F^{\times\,2}=\alpha F^{\times\,2} , and \kappa_{\alpha}(\beta)=\kappa_{\delta}(\beta) for all elements \beta\in F^{\times} coprime to \alpha\delta . ∎ Examples. \begin{array}[]{c|ccc|c}\raisebox{0.0pt}[11.19443pt][4.30554pt]{\rule{0.0pt}{8.61108pt}}F&r+s&\rho^{+}-\rho&\#{\operatorname{Kr}}_{4}(F)&\text{generators of ${\operatorname{Kr}}_{4}(F)$}\\ \hline\cr\raisebox{0.0pt}[11.19443pt][4.30554pt]{\rule{0.0pt}{8.61108pt}}{\mathbb{Q}}(i)&1&0&1&i\\ \raisebox{0.0pt}[11.19443pt][4.30554pt]{\rule{0.0pt}{8.61108pt}}{\mathbb{Q}}(\sqrt{2}\,)&2&0&2&-1,1+\sqrt{2}\\ \raisebox{0.0pt}[11.19443pt][4.30554pt]{\rule{0.0pt}{8.61108pt}}{\mathbb{Q}}(\sqrt{3}\,)&2&1&1&2+\sqrt{3}\\ \raisebox{0.0pt}[11.19443pt][4.30554pt]{\rule{0.0pt}{8.61108pt}}{\mathbb{Q}}(\sqrt{10}\,)&2&0&2&-1,3+\sqrt{10}\\ \raisebox{0.0pt}[11.19443pt][4.30554pt]{\rule{0.0pt}{8.61108pt}}{\mathbb{Q}}(\sqrt{15}\,)&2&1&1&4+\sqrt{15}\\ \raisebox{0.0pt}[11.19443pt][4.30554pt]{\rule{0.0pt}{8.61108pt}}{\mathbb{Q}}(\sqrt{39}\,)&2&1&1&8+\sqrt{39}\end{array} Observe that F={\mathbb{Q}}(\sqrt{39}\,) is an example of a real quadratic number field with E_{F}=E_{4} since -1\equiv\sqrt{39}^{2}\bmod 4 and the fundamental unit \varepsilon=25+4\sqrt{39} satisfies \varepsilon\equiv 1\bmod 4 . Ray class groups Before we give a proof of Theorem 4.1 we recall a few basic facts about ray class groups. The ray class group modulo {\mathfrak{m}} is defined by the exact sequence \begin{CD}1@>{}>{}>P_{\mathfrak{m}}/P_{\mathfrak{m}}^{1}@>{}>{}>I_{\mathfrak{m}}/P_{\mathfrak{m}}^{1}@>{}>{}>I_{\mathfrak{m}}/P_{\mathfrak{m}}@>{}>{}>1\end{CD} Here I_{\mathfrak{m}} is the group of fractional ideals coprime to {\mathfrak{m}} and P_{\mathfrak{m}} its subgroup of principal ideals in I_{\mathfrak{m}} ; the ray P_{\mathfrak{m}}^{1} consists of all ideals (\alpha) generated by elements \alpha\equiv 1\bmod{\mathfrak{m}} (congruence in the multiplicative sense). Observe that I_{\mathfrak{m}}/P_{\mathfrak{m}}\simeq\operatorname{Cl}(F) since every ideal class contains ideals coprime to {\mathfrak{m}} , and that I_{\mathfrak{m}}/P_{\mathfrak{m}}^{1}\simeq\operatorname{Cl}_{F}\{{\mathfrak{m}}\} by definition. Below we will use the abbreviation R_{\mathfrak{m}}=P_{\mathfrak{m}}/P_{\mathfrak{m}}^{1} . The connection with units is provided by the exact sequence \begin{CD}1@>{}>{}>E/E_{\mathfrak{m}}@>{}>{}>({\mathcal{O}}_{F}/{\mathfrak{m}})^{\times}@>{}>{}>R_{\mathfrak{m}}@>{}>{}>1,\end{CD} where E_{\mathfrak{m}} is the group of units \varepsilon\equiv 1\bmod{\mathfrak{m}} . These sequences imply the well known formula \#\operatorname{Cl}_{F}\{{\mathfrak{m}}\}=h\cdot\frac{\Phi({\mathfrak{m}})}{(E:E_{\mathfrak{m}})}. If we now take {\mathfrak{m}}=(4) and mod out squares (observing that E^{2}\subseteq E_{4} ), we obtain the exaxct sequences (2) \displaystyle 1\longrightarrow E/E_{4}\longrightarrow M_{4}/M_{4}^{2}\longrightarrow R_{4}/R_{4}^{2}\longrightarrow 1 (3) \displaystyle 1\longrightarrow R_{4}/R_{4}^{2}\longrightarrow\operatorname{Cl}_{F}\{{\mathfrak{m}}\}/\operatorname{Cl}_{F}\{{\mathfrak{m}}\}^{2}\longrightarrow\operatorname{Cl}(F)/\operatorname{Cl}(F)^{2}\longrightarrow 1. 4.1. Proof of the Main Theorem The main step in the proof of Thm. 4.1 is verifying the exactness of the first two rows of the commutative diagram in Figure 1 . \begin{CD}111\\ @V{}V{}V@V{}V{}V@V{}V{}V\\ 1@>{}>{}>E/E_{4}@>{\iota_{1}}>{}>{\operatorname{Kr}}(F)@>{\pi_{1}}>{}>P_{F}/P_{F}^{2}@>{}>{}>1\\ @V{}V{}V@V{}V{}V@V{}V{}V\\ 1@>{}>{}>M_{4}/M_{4}^{2}@>{\iota_{2}}>{}>{\operatorname{Dir}}(F)@>{\pi_{2}}>{}>I_{F}/I_{F}^{2}@>{}>{}>1\\ @V{}V{}V@V{}V{}V@V{}V{}V\\ 1@>{}>{}>R_{4}/R_{4}^{2}@>{}>{}>\mbox{\rm SCl}(F)@>{}>{}>\operatorname{Cl}(F)/\operatorname{Cl}(F)^{2}@>{}>{}>1\\ @V{}V{}V@V{}V{}V@V{}V{}V\\ 111\end{CD} Table 1. Fundamental diagram for the determination of \mbox{\rm SCl}(F) . Here • P_{F} is the group of nonzero principal ideals; • I_{F} is the group of nonzero fractional ideals; The vertical sequences essentially are the definitions of the groups at the bottom. The snake lemma then implies that the sequence in the third row is exact. This implies that \mbox{\rm SCl}(F) has the same order as \operatorname{Cl}_{F}\{4\}/\operatorname{Cl}_{F}\{4\}^{2} , and hence is isomorphic to this group since both are elementary abelian. Exactness of the top row We begin by verifying that the sequence \begin{CD}1@>{}>{}>E/E_{4}@>{\iota_{1}}>{}>{\operatorname{Kr}}(F)@>{\pi_{1}}>{}>P_{F}/P_{F}^{2}@>{}>{}>1\end{CD} is exact. The homomorphism \iota_{1} sends \varepsilon E_{4} to the Kronecker character \kappa_{\varepsilon} . It is well defined since for units \varepsilon\in E_{4} , \kappa_{\varepsilon} is the principal character. The homomorphism \pi_{1} sends a Kronecker character \kappa_{\alpha} to the coset (\alpha)P_{F}^{2} , where P_{F} is the group of nonzero principal ideals in F . In particular, \operatorname{im}\iota_{1}\subseteq\ker\pi_{1} . Conversely assume that \kappa is a Kronecker character in the kernel of \pi ; then (\alpha) is the square of a principal ideal, hence \alpha=\varepsilon\beta^{2} for a unit \varepsilon\in E and some \beta\in F^{\times} . Thus \kappa_{\alpha}=\kappa_{\varepsilon} , hence \kappa\in\operatorname{im}\iota_{1} . Exactness of the middle row Nest we prove the exactness of \begin{CD}1@>{}>{}>M_{4}/M_{4}^{2}@>{\iota_{2}}>{}>{\operatorname{Dir}}(F)@>{\pi_{2}}>{}>I_{F}/I_{F}^{2}@>{}>{}>1.\end{CD} The map \iota_{2} is defined as in Prop. 4.2 ; recall that \iota_{2} is injective, that it makes the left upper square commutative, and that \operatorname{im}\iota_{2}={\operatorname{Dir}}_{4}(F) . Let \chi be a primitive quadratic Dirichlet character with conductor {\mathfrak{m}} ; then we set \pi_{2}(\chi)={\mathfrak{m}}I_{F}^{2} . We claim that \pi_{2} is surjective. Clearly every prime ideal with odd norm is the conductor of a primitive Dirichlet character. If {\mathfrak{l}} is a prime ideal above 2 , choose an \alpha divisible exactly by {\mathfrak{l}} , and take the {\mathfrak{l}} -part of the Kronecker symbol \kappa_{\alpha} . Since I_{F}/I_{F}^{2} is generated by prime ideals, \pi_{2} is onto. If \chi\in\ker\delta , then the conductor of \chi is the square of an ideal. Since \chi is primitive, this implies that \chi is defined modulo an ideal dividing 4 , hence it is in the image of \iota_{2} by Prop. 4.2 . Examples In Table 2 we have listed a few examples of separant class groups for quadratic number fields with small discriminant. \begin{array}[]{c|c|c|c|c|c}\raisebox{0.0pt}[11.19443pt][4.30554pt]{\rule{0.0pt}{8.61108pt}}F&E_{4}/E^{2}&M_{4}/M_{4}^{2}&\operatorname{Cl}_{2}(F)&\mbox{\rm SCl}(F)&\text{generators of $\mbox{\rm SCl}(F)$}\\ \hline\cr\hline\cr\raisebox{0.0pt}[11.19443pt][4.30554pt]{\rule{0.0pt}{8.61108pt}}{\mathbb{Q}}&1&1&1&1&\\ \raisebox{0.0pt}[11.19443pt][4.30554pt]{\rule{0.0pt}{8.61108pt}}{\mathbb{Q}}(\sqrt{-1}\,)&1&{\mathbb{Z}}/2{\mathbb{Z}}&1&{\mathbb{Z}}/2{\mathbb{Z}}&\chi_{2i}\\ \raisebox{0.0pt}[11.19443pt][4.30554pt]{\rule{0.0pt}{8.61108pt}}{\mathbb{Q}}(\sqrt{-5}\,)&{\mathbb{Z}}/2{\mathbb{Z}}&({\mathbb{Z}}/2{\mathbb{Z}})^{2}&{\mathbb{Z}}/2{\mathbb{Z}}&({\mathbb{Z}}/2{\mathbb{Z}})^{2}&\chi_{2},\chi_{3}\\ \raisebox{0.0pt}[11.19443pt][4.30554pt]{\rule{0.0pt}{8.61108pt}}{\mathbb{Q}}(\sqrt{2}\,)&1&&1&1&\\ \raisebox{0.0pt}[11.19443pt][4.30554pt]{\rule{0.0pt}{8.61108pt}}{\mathbb{Q}}(\sqrt{3}\,)&{\mathbb{Z}}/2{\mathbb{Z}}&{\mathbb{Z}}/2{\mathbb{Z}}&1&{\mathbb{Z}}/2{\mathbb{Z}}&\chi_{2}\\ \raisebox{0.0pt}[11.19443pt][4.30554pt]{\rule{0.0pt}{8.61108pt}}{\mathbb{Q}}(\sqrt{6}\,)&{\mathbb{Z}}/2{\mathbb{Z}}&{\mathbb{Z}}/2{\mathbb{Z}}&1&{\mathbb{Z}}/2{\mathbb{Z}}&\chi_{4}\\ \raisebox{0.0pt}[11.19443pt][4.30554pt]{\rule{0.0pt}{8.61108pt}}{\mathbb{Q}}(\sqrt{10}\,)&1&1&{\mathbb{Z}}/2{\mathbb{Z}}&{\mathbb{Z}}/2{\mathbb{Z}}&\chi_{3}\\ \raisebox{0.0pt}[11.19443pt][4.30554pt]{\rule{0.0pt}{8.61108pt}}{\mathbb{Q}}(\sqrt{15}\,)&{\mathbb{Z}}/2{\mathbb{Z}}&{\mathbb{Z}}/2{\mathbb{Z}}&{\mathbb{Z}}/2{\mathbb{Z}}&({\mathbb{Z}}/2{\mathbb{Z}})^{2}&\chi_{2},\chi_{3}\\ \raisebox{0.0pt}[11.19443pt][4.30554pt]{\rule{0.0pt}{8.61108pt}}{\mathbb{Q}}(\sqrt{34}\,)&{\mathbb{Z}}/2{\mathbb{Z}}&{\mathbb{Z}}/2{\mathbb{Z}}&{\mathbb{Z}}/2{\mathbb{Z}}&{\mathbb{Z}}/2{\mathbb{Z}}&\chi_{3}\end{array} Table 2. Separant class groups \mbox{\rm SCl}(F) of a few number fields F . • F={\mathbb{Q}}(\sqrt{-1}\,) . Here the Dirichlet character with conductor 2 is not induced from a Kronecker character. • F={\mathbb{Q}}(\sqrt{-5}\,) . Here -1\equiv\sqrt{-5}^{2}\bmod 4 , hence E_{4}=\langle-1\rangle . The only squarefree elements whose norm is a power of 2 are -1 and \pm 2 ; the conductors of the Kronecker symbols (\frac{\pm 2}{\cdot}) are 4 , hence there is no Kronecker character with conductor 2 . There is also no Kronecker character with conductor {\mathfrak{p}} or {\mathfrak{p}}^{\prime} , the two prime ideals above 3 , since there is no quadratic extension ramified exactly at {\mathfrak{p}} . The product of the Dirichlet characters \chi_{\mathfrak{p}} and \chi_{{\mathfrak{p}}^{\prime}} is the Kronecker character (\frac{-3}{\cdot}) , hence these two characters are equivalent in \mbox{\rm SCl}(F) . • F={\mathbb{Q}}(\sqrt{3}\,) . We have -1\equiv\sqrt{3}^{2}\bmod 4 , hence E_{4}=\langle-1\rangle . We have the following Kronecker characters (\frac{\alpha}{\cdot}) whose conductors divide 4 : \begin{array}[]{c|ccccccc}\raisebox{0.0pt}[11.19443pt][4.30554pt]{\rule{0.0pt}{8.61108pt}}\alpha&-1&\varepsilon&-\varepsilon&\pi&-\pi&\varepsilon\pi&-\varepsilon\pi\\ \hline\cr\raisebox{0.0pt}[11.19443pt][4.30554pt]{\rule{0.0pt}{8.61108pt}}\operatorname{cond}\,\kappa_{\alpha}&\infty&(4)&(4)\infty&(4\pi)\infty_{1}&(4\pi)\infty_{2}&(4\pi)\infty_{1}&(4\pi)\infty_{2}\end{array} Here \varepsilon=2+\sqrt{3} is the fundamental unit of F and \pi=1+\sqrt{3} generates the prime ideal above 2 . Observe that 2=\pi^{2}/\varepsilon . In particular, there is no Kronecker character with conductor (2) . • F={\mathbb{Q}}(\sqrt{6}\,) . Here the fundamental unit is \varepsilon=5+2\sqrt{6} , and -\varepsilon\in E_{4} . The Kronecker character (\frac{-1}{\cdot})=(\frac{\varepsilon}{\cdot}) has conductor (2) . The Dirichlet character \chi_{4} with conductor 4 defined as the 2 -part of (\frac{3+\sqrt{6}}{\cdot}) is not a Kronecker character since there is no quadratic extension K/F with relative discriminant (3+\sqrt{6}\,) . • F={\mathbb{Q}}(\sqrt{10}\,) . The Kronecker character (\frac{-1}{\cdot}) has conductor (2) , and (\frac{1+\sqrt{10}}{\cdot}) has conductor (4) . The Dirichlet character \chi_{3} with conductor 3 is not a Kronecker character. • F={\mathbb{Q}}(\sqrt{15}\,) . The unit -1 generates E_{4}/E^{2} . If \varepsilon=4+\sqrt{15} denotes the fundamental unit, then \kappa_{\varepsilon} has conductor (4) . The Dirichlet characters with conductors 2 and 3 are not Kronecker characters, and neither is their product. • F={\mathbb{Q}}(\sqrt{34}\,) . The fundamental unit \varepsilon=35+6\sqrt{34} generates E_{4}/E^{2} . The Dirichlet character with conductor 3 is not a Kronecker character. 5. Separants Relative discriminants \operatorname{disc}\,(K/F) of quadratic extensions K/F of number fields contain information about the ramified primes. Separants, which we will define below, contain more information: they are (cosets of) algebraic numbers whose square roots generate the quadratic extensions in question. In this section we will define separants for number fields F with odd class number, and we will prove their most basic properties. For quadratic extensions K/F , we let \operatorname{Ram}(K/F) denote the set of ramified places of F , and \operatorname{Ram}_{f}(K/F) its subset of finite ramified places, i.e., of prime ideals in F that are ramified in K/F . Proposition 5.1 . Let K/F be a quadratic extension of number fields. If F has odd class number and K=F(\sqrt{\alpha}\,) , then there exists an integer \delta\in F^{\times} unique up to squares of units such that • F(\sqrt{\alpha}\,)=F(\sqrt{\delta}\,) ; • \operatorname{disc}\,(K/F)^{h}=(\delta) . Proof. Assume that F has odd class number h , and consider quadratic extensions K=F(\sqrt{\alpha}\,) . The relative discriminant {\mathfrak{d}}=\operatorname{disc}\,(K/F) is an ideal, and we know that 4\mu=\operatorname{disc}\,(\sqrt{\mu}\,)={\mathfrak{a}}^{2}{\mathfrak{d}} for some integral ideal {\mathfrak{a}} . Raising everything to the h -th power we get (4^{h}\mu^{h})=(\alpha^{2}\delta) , where {\mathfrak{a}}^{h}=(\alpha) and {\mathfrak{d}}^{h}=(\delta) ; moreover we can (and will) choose \delta in such a way that F(\sqrt{\mu}\,)=F(\sqrt{\delta}\,) . ∎ The coset {\operatorname{sep}}(K/F)=\delta\cdot E_{F}^{2} is called the separant of the quadratic extension K/F . Separants form a group \mbox{\rm Sep}(F) with respect to the following multiplication: if K_{1}=F(\sqrt{\alpha}\,) and K_{2}=F(\sqrt{\beta}\,) are quadratic extensions with separants \delta_{1}E_{F}^{2} and \delta_{2}E_{F}^{2} , let K_{3}=F(\sqrt{\alpha\beta}\,) be the third quadratic subfield of K_{1}K_{2}/F , and set \delta_{1}E_{F}^{2}*\delta_{2}E_{F}^{2}=\delta_{3}E_{F}^{2}, where \delta_{3}E_{F}^{2}={\operatorname{sep}}(K_{3}/F) . Observe that this is the usual product if and only if the separants are coprime. Over the field F={\mathbb{Q}} , a separant of a quadratic number field is just its discriminant because E_{\mathbb{Q}}^{2}=1 . Example 2. Consider F={\mathbb{Q}}(\sqrt{-23}\,) and K=F(\sqrt{\mu}\,) for \mu=-5+2\sqrt{-23} . Since (\mu)={\mathfrak{p}}^{2}{\mathfrak{q}} for the prime ideals {\mathfrak{p}}=(3,1-\sqrt{-23}\,) and {\mathfrak{q}}=(13,4+\sqrt{-23}\,) , we have \operatorname{Ram}_{f}(K/F)=\{{\mathfrak{q}}\} and \operatorname{disc}\,K/F={\mathfrak{q}} . Also we find {\mathfrak{q}}^{3}=(\delta) for \delta=-37-6\sqrt{-23} , and {\operatorname{sep}}(K/F)=\delta E_{F}^{2} . Separants that cannot be written as a product of separants are called irreducible; separants \delta\cdot E_{F}^{2} for which (\delta) is a power of a prime ideal are called prime separants. Example 3. Consider the quadratic extension K=F(\sqrt{1+2i}\,) of F={\mathbb{Q}}(i) . It has relative integral basis \{1,\frac{1+\sqrt{1+2i}}{1+i}\} , and its relative discriminant is therefore the ideal (1+i)^{2}(1+2i) . The generator \delta=2i(1+2i) of this ideal has the property that K=F(\sqrt{\delta}\,) , hence it is the separant of this extension: {\operatorname{sep}}(K/F)=2i(1+2i) . Since there is no quadratic extension of F with relative discriminant {\mathfrak{d}}=(2) , this separant cannot be factored into prime separants. The modulus associated to a separant {\operatorname{sep}}(K/F) is the conductor of the extension K/F , including the infinite places; two separants are called coprime if they are coprime as ideals, and strongly coprime if their associated conductors are coprime. Since the square root of the separant {\operatorname{sep}}(K/F) generates the quadratic extension K/F , the splitting of primes in K/F is determined by the quadratic Kronecker symbol (\delta/{\mathfrak{p}}) . 6. Unique Factorization into Prime Separants We now explain how to generalize the prime discriminant factorization considered by Goldstein and Sunley to fields with trivial separant class group: Theorem 6.1 . Let F be an algebraic number field. Then the following assertions are equivalent: (1) The separant class group of F is trivial: \mbox{\rm SCl}(F)=1 . (2) F is totally real, and its class number in the strict sense is odd. (3) F has odd class number, and for every Dirichlet character \chi defined modulo 4 there is a unit \varepsilon\in E_{F} such that \chi=(\frac{\varepsilon}{\cdot}) . (4) F has odd class number, and for each prime ideal {\mathfrak{p}} with odd norm there is a (unique) quadratic extension K/F with \operatorname{Ram}_{f}(K/F)=\{{\mathfrak{p}}\} . (5) Every separant in \mbox{\rm Sep}(F) can be written uniquely (up to order) as a product of prime separants. (6) For every quadratic extension K/F we have (K_{\operatorname{gen}}^{(2)+}:F)=2^{t} , where t=\#\operatorname{Ram}_{f}(K/F) . Proof. (1) \iff (2) is a consequence of the formula \#\mbox{\rm Sep}(F)=2^{\rho^{+}+s} , where \rho^{+}={\operatorname{rk}\,}\operatorname{Cl}_{2}(F) and 2s is the number of complex embeddings of F . (1) \Longrightarrow (4): Let {\mathfrak{p}} be a prime ideal with odd norm. Since F has odd class number, {\mathfrak{p}}^{h}=(\pi) for some \pi\in{\mathcal{O}}_{F} . Since \mbox{\rm SCl}(F)=1 , every element of M_{4}/M_{4}^{2} is reprsented by a unit; thus there is a unit \varepsilon such that \varepsilon\pi\equiv\xi^{2}\bmod 4 . In the quadratic extension K=F(\sqrt{\pi\varepsilon}\,) , the only ramified prime ideal is {\mathfrak{p}} . (4) \Longrightarrow (3): Let \chi be a Dirichlet character defined modulo 4 . Then \chi is in the image of M_{4}/M_{4}^{2} , i.e. there is a Kronecker character \kappa=(\frac{\alpha}{\cdot}) such that \kappa=\chi\psi , where \psi is defined modulo an odd ideal. By assumption, \psi is a Kronecker character; but then so is \chi . This implies the claim. (3) \Longrightarrow (1): By Cor. 4.5 F is totally real and \rho^{+}=\rho=0 . (1) \Longrightarrow (5): Let \delta E_{F}^{2} be a separant. The Kronecker character \kappa_{\delta} can be written uniquely as a product of primitive quadratic Dirichlet characters with prime power conductor: \kappa_{\delta}=\chi_{1}\cdots\chi_{t} . Since \mbox{\rm Sep}(F)=1 , each factor \chi_{j} is a Kronecker character (\delta_{j}/\,\cdot\,) . Since the conductors of these Kronecker characters are coprime prime ideal powers we must have \delta=\delta_{1}\cdots\delta_{t} . We claim that this factorization is unique. To this end, assume that there are two factorizations \delta=\delta_{1}\cdots\delta_{t}=\delta_{1}^{\prime}\cdots\delta_{t}^{\prime} of \delta into prime separants. Since K_{\operatorname{gen}}^{+}=F(\sqrt{\delta_{1}},\ldots,\sqrt{\delta_{t}}\,) is the maximal elementary abelian extension of F unramified over K at all finite primes, we must have \sqrt{\delta_{j}^{\prime}}\in K_{\operatorname{gen}}^{+} . But since \delta_{j}^{\prime} and the \delta_{j} are prime separants, this is only possible if \delta_{j}^{\prime}E^{2} is among the \delta_{j}E^{2} . (5) \Longrightarrow (6): Let K/F be a quadratic extension with separant \delta . Since \mbox{\rm SCl}(F)=1 we can write \delta=\delta_{1}\cdots\delta_{t} as a product of prime discriminants that are coprime except for infinite prime factors. But then L=F(\sqrt{\delta_{1}},\ldots,\sqrt{\delta_{t}}\,)/K is unramified at all finite primes, hence equal to the genus field in the strict sense. (6) \Longrightarrow (4): We first claim that F has odd class number in the strict sense. If not, then there is a quadratic extension K/F unramified at all primes; in particular, t=0 , and by assumption we have K_{\operatorname{gen}}^{+}=F : contradiction since K\subseteq K_{\operatorname{gen}}^{+} . Now let {\mathfrak{p}} be a prime ideal with odd norm; choose a prime ideal {\mathfrak{q}} in the inverse class of {\mathfrak{p}} with odd norm in the ray class group \operatorname{Cl}\{4\} modulo 4 ; then {\mathfrak{p}}{\mathfrak{q}}=(\alpha) for some \alpha\equiv\xi^{2}\bmod 4 . The quadratic extension K=F(\sqrt{\alpha}\,) has t=2 ramified prime ideals, hence K_{\operatorname{gen}}^{+}=F(\sqrt{\delta}_{1},\sqrt{\delta_{2}}\,) for coprime separants \delta_{1} , \delta_{2} . One of these separants generates a quadratic extension in which only {\mathfrak{p}} ramifies. ∎ 7. Genus theory Genus theory of quadratic extensions k/F of number fields with trivial separant class group can be made as explicit as over {\mathbb{Q}} : the factorization of \delta={\operatorname{sep}}(k/F) into prime separants gives us the genus class field as well as the genus characters. Lemma 7.1 . Let F be a number field with odd class number in the strict sense. Let K=F(\sqrt{\delta}\,) and L=F(\sqrt{\delta_{1}},\sqrt{\delta_{2}}\,) , were \delta , \delta_{1} and \delta_{2} are separants. Then L/K is unramified at all finite primes if and only if \delta=\delta_{1}\delta_{2} . Proof. Let k_{1}=F(\sqrt{\delta_{1}}\,) , k_{2}=k(\sqrt{\delta_{2}}\,) and K=F(\sqrt{\delta_{1}\delta_{2}}\,)=F(\sqrt{\delta}\,) be the three quadratic subextensions of K/F . By the conductor-discriminant formula we have \operatorname{disc}\,(L/F)=\operatorname{disc}\,(k/F)\cdot\operatorname{disc}\,(k_{1}/F)\cdot\operatorname{disc}\,(k_{2}/F). Since L/K is unramified, we have \operatorname{disc}\,(L/F)=\operatorname{disc}\,(k/F)^{2} . This implies \operatorname{disc}\,(k_{1}/F)\cdot\operatorname{disc}\,(k_{2}/F)=\operatorname{disc}\,(k/F) . Since separants are h -th powers of the relative discriminants, we must have \delta_{1}\cdot\delta=\varepsilon\delta . Since k=F(\sqrt{\delta_{1}\delta_{2}}\,)=F(\sqrt{\delta}\,) , the unit \varepsilon must be a square, and this implies that \delta_{1}E^{2}\cdot\delta_{2}E^{2}=\delta E^{2} as claimed. ∎ Let F be a number field with \mbox{\rm SCl}(F)=1 . For a quadratic extension K/F with separant {\operatorname{sep}}(K/F)=\delta E_{F}^{2} let \delta=\delta_{1}\cdots\delta_{t} denote the factorization of {\operatorname{sep}}(K/F) into prime separants. We already know that this factorization determines the genus field of K/F : Proposition 7.2 . Let F be a number field with trivial separant class group, and let K/F be a quadratic extension with separant \delta={\operatorname{sep}}(K/F) . If \delta=\delta_{1}\cdots\delta_{t} is its factorization into prime separants, then its genus field in the strict sense is K_{\operatorname{gen}}^{+}=F(\sqrt{\delta_{1}},\ldots,\sqrt{\delta_{t}}\,) . Proof. Write {\operatorname{sep}}(k/F)=\delta=\delta_{1}\cdots\delta_{t} as a product of prime separants. Then K=F(\sqrt{\delta_{1}},\ldots,\sqrt{\delta_{t}}\,) is an abelian extension of F unramified over k , hence contained in the genus field. Since (k_{\operatorname{gen}}:k)\leq 2^{t} , we must have K=k_{\operatorname{gen}} . ∎ Now let c\in\operatorname{Cl}(K) be an ideal class, pick an ideal {\mathfrak{a}}\in c coprime to \delta . Then \chi_{j}(c)=(\frac{\delta_{j}}{N_{K/F}{\mathfrak{a}}}) is a well defined quadratic character on the class group. The characters \chi_{j} are called genus characters of K/F . Theorem 7.3 (Principal Genus theorem) . Let F be a number field with trivial separant class group. Then an ideal class c\in\operatorname{Cl}^{+}(K) is a square if and only if \chi_{j}(c)=+1 for all genus characters \chi_{j} . Proof. Write c=[{\mathfrak{p}}] for some prime ideal {\mathfrak{p}} unramified in k/F . Clearly \chi_{j}(c)=1 for all genus characters is equivalent to {\mathfrak{p}} splitting completely in the genus class field. Since F has odd class number in the strict sense, K_{\operatorname{gen}}^{+} is the maximal elementary abelian 2 -extension contained in the Hilbert class field. By class field theory, {\mathfrak{p}} splits completely in the maximal elementary abelian 2 -extension containd in the Hilbert class field if and only if the ideal class [{\mathfrak{p}}] is a square in the class group \operatorname{Cl}_{2}^{+}(k) . ∎ Example. Consider the quadratic extension k=F(\sqrt{93+50\sqrt{2}}\,) of F={\mathbb{Q}}(\sqrt{2} . The separant \delta={\operatorname{sep}}(k/F)=(107+55\sqrt{2}\,)E_{F}^{2} has the factorization into prime separants 107+54\sqrt{2}=(13+4\sqrt{2}\,)(7+2\sqrt{2}\,). This shows that the genus class field of k is given by k_{\operatorname{gen}}^{+}=k(\sqrt{13+4\sqrt{2}}\,). Since k_{\operatorname{gen}}^{+}=k_{\operatorname{gen}} in this case, the 2 -class group of k must be cyclic. The 2 -class group is generated by a prime ideal above 2 ; since (\sqrt{2}\,) does not split in F(\sqrt{13+4\sqrt{2}}\,) , the ideal cass generated by the prime ideal above 2 is not a square, and we deduce that \operatorname{Cl}_{2}(k)\simeq{\mathbb{Z}}/2{\mathbb{Z}} . Final Remarks Separants were used in [ 8 ] for constructing unramified 2 -extensions of quadratic extensions K/F of fields F with \mbox{\rm SCl}(F)=1 . I expect that the main results of Rédei (Rédei matrix, triple symbol reciprocity; see Stevenhagen [ 11 ] ), Koch [ 7 ] , Fröhlich [ 4 ] and the existence of governing fields generalize to all number fields with trivial separant class group.